Authority Receipts for AI-Assisted Insurance
MAND8 generates Authority Receipts that make AI-assisted insurance actions provable, reviewable, and insurable.
Thesis infographic
MAND8 — decision instrument
Recombination equation
Assessment radar
advanceGating axes (bold) require ≥ 4 to advance from Genesis selection.
Buyer → blocker → evidence
Economic buyer
Blockers
Required evidence
90-day pilot
Can shadow-mode Authority Receipts, generated from existing workflow events over 90 days at one managing agent, reconstruct authority for every AI-assisted action to a DA reviewer's satisfaction?
Day 0
Kickoff
Day 30
20 DA operator interviews complete. Authority Receipt schema v0.1 published. Workflow boundary written.
Day 60
Shadow pilot design signed with one managing agent. Regulatory heat-map reviewed by external counsel. Validator prototype running against synthetic data.
Day 90
90-day shadow pilot in flight. Economic buyer named with budget line. Evidence audit scheduled with DA expert, Lloyd's syndicate, and claims-ops reviewer.
Success metrics
- ✓95% Authority Receipt completeness across in-scope actions.
- ✓100% of high-risk actions capture human review state.
- ✓Evidence for any single action reconstructable in minutes, not weeks.
Failure criteria
- ✕DA reviewer cannot use receipts to answer the four authority questions.
- ✕Buyer will not fund a follow-on pilot at commercial rate.
- ✕Receipts duplicate an existing audit-log field with no additional signal.
UK regulatory heat
Decision
keepThesis survives Week 0. One primitive (Authority Receipt), one workflow (AI-assisted delegated authority), one measurable pilot, one plausible economic buyer. Commercial evidence remains the weakest score — the next 12 weeks exist to generate it or kill the thesis.
Next artifact change: Move from thesis document to Week 1 buyer-discovery interview plan (20 named DA operators).
- Buyer discovery interview plan — Bankabil · Week 1
- Receipt schema v0.1 (JSON) — Bankabil / BLKBXS · Week 3
- Shadow pilot design memo — Live Oak Bank Labs evidence lead · Week 4
One-sentence thesis
MAND8 enters Live Oak Bank Labs as a Genesis Studio thesis. The goal of the next weeks is not to build a product — it is to accumulate evidence, or kill the thesis.
Analogy
Delegated authority is the one insurance workflow where authority is already explicit — every managing agent already carries a binder that says who is allowed to bind what. An Authority Receipt is what proves an AI-assisted action stayed inside that binder.
Per the Live Oak Bank Labs Genesis schema, MAND8 enters the lab as a Genesis Studio thesis, not a startup. Every week produces buyer, regulatory, pilot, diligence, or kill evidence. The strongest objection — nobody buys "authority" — is the whole reason the thesis is in the lab.
Cohort fit — lineage recombination
| Primitive | Translation |
|---|---|
| Lloyd's Delegated Authority | Binder-level authority is already explicit, coded, and audited. MAND8 rides on top of an existing operating primitive. |
| nCino Workflow | Bank workflow standardization applied to authority validation instead of credit review. |
| BLKBXS Receipts | Tamper-evident evidence bundles as the underlying infrastructure primitive. |
| Operational Risk | Authority breaches are already an op-risk category — MAND8 turns them into a measurable one. |
| AI / Model Governance | Existing model-risk vocabulary (approval, review, threshold) maps directly to authority states. |
Genesis scorecard
| Dimension | Score | Rationale |
|---|---|---|
| Buyer urgency | 4.5 | AI governance is becoming operational, but buyer validation is still needed. |
| Product clarity | 4.5 | "Authority Receipt" is a crisp primitive — one record, eight fields, one workflow. |
| Regulatory tractability | 4 | Operates as evidence infrastructure; does not make insurance or regulatory claims. |
| Technical defensibility | 5 | BLKBXS receipt infrastructure plus an authority model form a differentiated foundation. |
| Pilot clarity | 5 | A shadow-mode delegated authority workflow is concrete, low-risk, and time-boxed to 90 days. |
| Commercial evidence | 2 | Weakest area. No validated design partners yet. The Genesis process exists to fix this or kill the thesis. |
Verdict. Advance to Genesis Week 1. Commercial evidence is the weakest link — the next weeks exist to generate or kill it.
Final product packaging
Primitive
Authority Receipt. A single record proving an AI-assisted action stayed inside a delegated authority — with policy, reviewer, approval state, and evidence hash.
Workflow
AI-assisted delegated authority. Not claims. Not underwriting. Not pricing. Authority is already explicit — the product naturally fits.
Pilot
90-day shadow pilot with one managing agent. Success = 95% receipt completeness, 100% high-risk review captured, evidence reconstructed in minutes.
MAND8 proves whether an AI-assisted insurance action was authorized before it happened.Enter to disprove. Every week ends in an artifact, a scored decision, and an updated risk ledger — or the thesis is killed.